Worked on the pytorch/executorch repository to deliver a Conv1d weight channel packing optimization, targeting improved memory efficiency and throughput for 1D convolution workloads. The approach involved restructuring how Conv1d weight tensors are stored and accessed, resulting in reduced memory footprint and faster convolution operations. Leveraging expertise in Convolutional Neural Networks, GPU programming, and Vulkan, the work focused on low-level tensor layout changes and performance profiling. All modifications were validated against representative workloads to ensure stability, with no regressions observed. This optimization enables more efficient training and inference for convolution-heavy models, potentially lowering hardware requirements and runtime for end users.